433% in 3 months - impressive?

[Almunecar]

I'm an almost newbie, lost too much money on spreadbetting. Decided there had to be a way of making a system that works. I've played around with backtesting before and never seemed to come up with anything worthwhile. Now I've hit upon something that APPEARS (?!) to bring a return. See the attached statistics chart. It represents 3 months trading on the Dow with a 10,000 account. If it were guaranteed I'd be home and dry, but I'm sure the whole thing is riddled with holes. The system runs on a 1 hour Dow chart, run it on a 15 min or 5min chart and the results are more painful.

I'm looking for someone that understand system creation and backtesting better than I do (Thats a BIG field!).

I'd like to work with them confidentially at tweaking what I've done (I can see possible ways of making improvements but don't know technically HOW to include the parameters in the system).

Any interest?

Mail me [email protected]

 

[shadowninja]

One question: how much % of your account are you risking per trade?

 

[A Dashing Blade]

. . It represents 3 months trading on the Dow . . .

Google "curve fitting" and "data mining"

 

[Hoggums]

Google "curve fitting" and "data mining"

Absolutely.

Next step is to take your "system" and test it on data over a longer period that you never tested with - say from 2001-2003 and see if it makes a profit.

My prediction is that it'll blow up.

 

[trendie]

Almunecar,

as per Hoggums, please test your rules against another 3-month period in 2007.
eg; Months 1-3, 2-4, 3-5, 4-6, etc, to get a feel if the results are representative of your posted results.

To avoid the possibility of curve-fitting, change one of your settings ever so slightly. If your rules are robust, the results should change slightly. If they change markedly, chances are its curve fitted.

Another test is to ensure that your results dont rely on a very small number of very large wins, ie, without them the performance drops.

Otherwise, I guess you're onto a winner.

EDIT: Curtis Faiths "Way of the Turtle" explains some methods to ensure robustness of rules, and the pitfalls of optimisation. (although he doesnt call it a bad idea itself)

 

[anley]

Not impressive because it shows the strategy has little respect for risk and risk, or rather controlling it is what this game is all about. Profits are a secondary consideration, but of course still very important

To make that sort of money over that time period can only mean one thing, you're taking far too much risk. This translates into a very simple fact, the chances of that startegy going bust in the future are 1.0, not 0.95, not 0.99 but 1.0

Tone it down and as others say test it over a longer period of time and over all sorts of data then it might be worthwhile. Also, watch your costs because if it's in and out all the time in realtime trading what you'll most probably find is that most of the profits get eaten up in costs........

 

[the blades]

Also, be very aware that most (if not all) mechanical systems don't work forever - take a look at collective 2 to see the number of backtested forex systems that do really well for a short period of time. Forex has only a few currency pairs with loads of traders developing mechanical systems - ensuring the market is fast changing.

That said - get more data and test your system in sample and out of sample on several periods moving forwards - if you can re-optimise the system and see not too much drop off out od sample, you have grounds for encouragement. Also, look at adding some noise to the data (monte carlo analysis) - if the performance drops off as you add more noise, there's a big risk you've over fitted your rules to the data. If not, again grounds for encouragement.

UTB

 

[theblackmamba]

just come back from almunecar.. love the place, apart from the building work on promenade

 

[timsk]

just come back from almunecar.. love the place, apart from the building work on promenade

Yeah, me too - like it that is. It's very 'earthy' and not too touristy. Salobrena's good too, which isn't a million miles away from Almunecar.

 

[forex999]

actually i could do with some help, I am looking to automate a system, whereby the following happens;

1. Prices crosses a moving average; 30 min, upside for long
2. After the cross the price retraces 50+% fib 15 min
3. After 2 above the price crosses a moving average on a 5 minute
4. MACD must be above the waterline on 5 minute

all the above must happen, and once confirmed a buy signal. I hate watching the charts but it produces a 70% accuracy

what can this be automated?
which charting package is best to programme?

any help!

 

[playhouse2005]

How did you get a 70% win rate, you based it on?
Just to let you know and from experience to fully automate a system is harder than the brain thinks. However, it can easily be done but ...
1) You will get much more signals then first thought.
2) You need to define it much better because there is no brain to operate it as it will
just follows the rules programmed and triggers automatically so if faulty you will very quickly lose ALL you account lol

Something to think about...
If you're very confident your system can produce above 70% winners then why not spend the time to automate it?
By the way 70% winners isn't much information. Does it produce 70% per decade? Per year?
What is the drawdowns? How often will it trigger a trade? How much will the contracts and slippage cost?
You have a lot to think about.

Feel free to share your thought, i'm more than happy to build it for you at a reasonable cost.

E